studying the dark triad of personality through twitter...
TRANSCRIPT
Studying the Dark Triad of Personalitythrough Twitter Behavior
Daniel Preotiuc-PietroJordan Carpenter, Salvatore Giorgi, Lyle Ungar
Positive Psychology CenterComputer and Information Science
University of Pennsylvania
October 26, 2016
Motivation
Online spaces are a medium for self-expression and socialcommunication.
There is a concern that these offer a medium for expressingdarker traits of human personality such as:
I Self-promotionI VanityI Anti-social behaviorI Alteration of the truthI Self-interest
The Dark Triad
The standard model in psychology for malevolent humanpersonality traits.I Coined in (Paulhus & Williams, 2002)
Assessed through questionnaires.I Similar to the ‘Big Five’ personality traits
Psychological studies on self-reported behaviors, notdata-driven exploration.I Social media offers a unique window into how people that demonstrate
these behaviors think and act
User Profiling
User profiling automatically quantifying traits from a user’sonline footprints:
I TextI ImagesI Platform usageI LikesI Social networkI ...
User Profiling
Two sides of the problem:
1. MeasurementI Goal: build models to predict traits of unknown usersI Predictive setup (regression/classification)I Using large scale Machine Learning
2. InsightI Goal: gain a better understanding of group differencesI Interpretable featuresI Use domain experts in analysis
Narcissism
Narcissism:
I VanityI EntitlementI Self-sufficiencyI SuperiorityI AuthorityI ExhibitionismI Exploitativeness
Sample Items:
I I tend to want others to admire me.I I tend to expect special favors from
others.
Narcissism
Miranda Priestly – The Devil Wears Prada
Psychopathy
Psychopathy:
I Lack of remorseI Lack of empathyI Pathological lyingI Need for stimulationI Superficial charmI Grandiose self-worth
Sample Items:
I I tend to lack remorse.I I tend to not be too concerned with
morality or the morality of my actions.
Psychopathy
Anton Chigurh – No Country for Old Men
Machiavellianism
Machiavellianism:
I DuplicitousI Ends justify the meansI Rarely reveal their true intentionsI Manipulate to get aheadI Money and power over relationshipsI FlatteryI Cynical view of human nature
Sample Items:
I I have used deceit or lied to get my way.I I tend to exploit others towards my
own end.
Machiavellianism
Frank Underwood – House of Cards
Data Set
Collected through a study on Amazon Mechanical Turk.
863 Twitter users with public profiles.
491 Twitter users posted > 500 tokens.
Collected all their tweets (<3200), their profile picture andprofile information.
Dark Triad Score
Completed the ’DirtyDozen’ questionnaire:I 12 questions;I 1–5 scale;I 4 questions/trait.
Reported age andgender.
We use the log of the traits for the restof the experiments.
Trait Inter-CorrelationI Treats are moderately
inter-correlated – asexpected;
I We compute an additional‘Dark Triad’ score as theaverage of the three inaccordance to previouswork;
I In our analysis of eachtrait, we control for theother two traits in additionto age and gender usingpartial correlation toisolate distinctivebehaviors.
Features – Text
I Unigrams:I Single tokens used by at least 10% of users (N = 6,491)
I LIWC:I Manually constructed word categories (Pennebaker et al,
2001)I Include parts-of-speech, topical categories, emotions (N =
64)I Topics:
I Obtained by using spectral clustering over word2vec wordrepresentations (Preotiuc-Pietro et al, 2015)
I Words that appear in similar contexts (N = 200)I Sentiment & Emotions:
I Messages tagged with either sentiment or discrete emotions(Mohammad et al. 2010)
I Each user is assigned its average message emotion scores(N = 10)
Features – Profile Image
I Color features:I Grayscale, Brightness, Contrast, Saturation, Sharpness, Blur
I Facial features:I Type of image: default, # faces, one face, multiple faces
(Face++)I Facial presentation: ratio, glasses, posture, smile
Features – Platform Usage
I Profile features:I No. tweets, tweets/dayI # friends, #followers, follower–friend ratio, #listedI Default background, geo-enabledI Proportion and count of tweets that were retweeted or liked
I Shallow features:I # characters, # tokens per tweetI Retweets or duplicate messagesI Proportion of messages which use hashtags, @-replies,
@-mentions, URLs or ask for followers
‘Core’ Dark Triad
Word2Vec Topics
R=.152 R=.126 R=.126 R=.117
Posting about work and addresses.
Topics significant at p<.01 (two-tailed t-test), controlled for Age and Gender.
‘Core’ Dark Triad
LIWC Categories
SWEARR=.127
ANGERR=.123
SPACER=.119
PRESENTR=.106
Related to present activities.
Topics significant at p<.01 (two-tailed t-test), controlled for Age and Gender.
‘Core’ Dark Triad
Emotions
NegativeR=.108
DisgustR=.102
TrustR=.093
Overall negative emotions, but also trust.
Topics significant at p<.01 (two-tailed t-test), controlled for Age and Gender.
‘Core’ Dark Triad
Image:
I less likely to be GrayscaleI lower sharpness
Profile:
I –
Shallow:
I Fewer characters per tweetI Fewer retweets performedI Fewer tweets with hashtags and URLs
All correlations significant at p<.05; controlled for age and gender.
Narcissism
Word2Vec Topics
R=.119 R=.111 R=.110 R=.104
Positive face to the world.
Support causes, celebrities, TV shows.
Post about their mundane activities on Twitter (which theythink others are interested in).
Topics significant at p<.01 (two-tailed t-test), controlled for Age, Gender,Machiavellianism and Psychopathy.
Narcissism
Emotions
R=.130Trust
R=.104Positive
Positive face to the world.
Positive emotions overlap in most frequent words.
Topics significant at p<.01 (two-tailed t-test), controlled for Age, Gender,Machiavellianism and Psychopathy.
Narcissism
Image:
I Not grayscaleI Prefer one face in profile image and not multiple facesI Smiling
Profile:
I Not default backgroundI Geo-enabledI More tweets that are favorited
Shallow:
I Fewer duplicate tweets (content curation)I Less tweets with hashtags and @-mentions
All correlations significant at p<.05; controlled for age, gender, psychopathyand Machiavellianism.
Psychopathy
Word2Vec Topics
R=.144
R=.116
R=.142
R=.110
R=.123
R=.110
R=.123
R=.108
Interested in news about violent activities and news (including‘Positive’ aggression), emergencies, issues.
Psychopathy
LIWC Categories
R=.153DEATH
R=.138ANGER
R=.110NEGEMO
R=.101BODY
Topics significant at p<.01 (two-tailed t-test), controlled for Age, Gender,Machiavellianism and Psychopathy.
Psychopathy
Emotions
R=.189Negative
R=.177Disgust
R=.174Fear
R=.173Anger
The entire spectrum of negative emotions.
Topics significant at p<.01 (two-tailed t-test), controlled for Age, Gender,Machiavellianism and Psychopathy.
Psychopathy
Image:
I Less saturated
Profile:
I –
Shallow:
I Fewer URLsI Not asking for followers
All correlations significant at p<.05; controlled for age, gender,Machiavellianism and narcissism.
Machiavellianism
Text:
I –
Image:
I –
Profile:
I Fewer retweetsI Fewer tweets with URLs
Shallow:
I –
All correlations significant at p<.05; controlled for age, gender, psychopathyand narcissism.
Prediction
.04
.01
.04
.10
.00
.05
.10
.15
.20
.25
Image
Narc Psyc Mach DT
Linear Regression, Pearson correlation between predictions andlog-scored traits, 10-fold cross-validation.
Prediction
.04
.09
.01 .01
.04
.00
.10
.05
.00
.05
.10
.15
.20
.25
Image Profile
Narc Psyc Mach DT
Linear Regression, Pearson correlation between predictions andlog-scored traits, 10-fold cross-validation.
Prediction
.04
.09
.14
.01 .01 .02
.04
.00
.12
.10
.05
.11
.00
.05
.10
.15
.20
.25
Image Profile Shallow
Narc Psyc Mach DT
Linear Regression, Pearson correlation between predictions andlog-scored traits, 10-fold cross-validation.
Prediction
.04
.09
.14
.16
.01 .01 .02 .02
.04
.00
.12
.20
.10
.05
.11
.16
.00
.05
.10
.15
.20
.25
Image Profile Shallow Emotions
Narc Psyc Mach DT
Linear Regression, Pearson correlation between predictions andlog-scored traits, 10-fold cross-validation.
Prediction
.04
.09
.14
.16.15
.01 .01 .02 .02
.14
.04
.00
.12
.20
.16
.10
.05
.11
.16 .16
.00
.05
.10
.15
.20
.25
Image Profile Shallow Emotions Unigrams
Narc Psyc Mach DT
Linear Regression, Pearson correlation between predictions andlog-scored traits, 10-fold cross-validation.
Prediction
.04
.09
.14
.16.15 .15
.01 .01 .02 .02
.14
.09
.04
.00
.12
.20
.16
.25
.10
.05
.11
.16 .16
.18
.00
.05
.10
.15
.20
.25
Image Profile Shallow Emotions Unigrams LIWC
Narc Psyc Mach DT
Linear Regression, Pearson correlation between predictions andlog-scored traits, 10-fold cross-validation.
Prediction
.04
.09
.14
.16.15 .15
.23
.01 .01 .02 .02
.14
.09
.21
.04
.00
.12
.20
.16
.25
.21
.10
.05
.11
.16 .16
.18.19
.00
.05
.10
.15
.20
.25
Image Profile Shallow Emotions Unigrams LIWC Topics
Narc Psyc Mach DT
Linear Regression, Pearson correlation between predictions andlog-scored traits, 10-fold cross-validation.
Prediction
.04
.09
.14
.16.15 .15
.23.25
.01 .01 .02 .02
.14
.09
.21
.25
.04
.00
.12
.20
.16
.25
.21
.25
.10
.05
.11
.16 .16
.18.19
.24
.00
.05
.10
.15
.20
.25
Image Profile Shallow Emotions Unigrams LIWC Topics All
Narc Psyc Mach DT
Linear Regression, Pearson correlation between predictions andlog-scored traits, 10-fold cross-validation.
Take Aways
I positive face to the worldI post about mundane
things
I profane and interest inviolence and violent events
I distinguished by fewerbehaviors beyond core‘Dark Triad’
Take Aways
I First data-driven approach exploring a core set ofpsychological traits (‘Dark Triad’)
I Multiple modalities: text, profile image and platformusages
I Text offers best predictive accuracy
I Predictive model of the dark triad traits from text publiclyreleased
Thank you!
www.preotiuc.ro